Introduction
Acute kidney injury (AKI) is a common serious complication of critically ill patients. AKI occurs in up to 50% patients in intensive care unit (ICU), with poor clinical prognosis [
1‐
4]. Patients with AKI are characteristic of a rapid loss of the kidney function, which can lead to electrolyte disorder, metabolic acidosis, fluid overload, and an increase in serum uremic toxins. Renal replacement therapy (RRT) has been widely used in critically ill patients with AKI. For patients with severe complications such as acute pulmonary edema, severe acidosis, and severe hyperkalemia, RRT is the cornerstone of AKI treatment to be performed urgently [
5,
6]. However, without these urgent indications, the optimal timing of initiating RRT is still under debate. Early initiation of RRT can correct metabolic disorders, control disturbances of fluid metabolism, and remove uremic toxins quickly and effectively. However, for patients whose renal function can recover spontaneously, early initiation of RRT may not be beneficial but expose them to the risk of RRT-associated adverse events, such as hemodynamic instability, bleeding, and bloodstream infection [
6‐
8].
Although there were several meta-analyses to evaluate whether critically ill patients with AKI can benefit from initiating RRT early, the conclusions were inconsistent and none of them included all randomized clinical trials (RCTs) up to present. Karvellas et al. conducted a meta-analysis, including two RCTs, four prospective cohort, and nine retrospective cohort, showing a beneficial impact on survival when RRT was performed at early stage [
9]. However, recently published meta-analyses on this topic indicated that early initiation of RRT did not improve patient prognosis [
10‐
12]. And a high-quality meta-analysis of RCTs with individual data of all the included patients reached the similar conclusion [
13]. Recently, the largest RCT, STARRT‑AKI trail, was published. Totally, 2927 critically ill patients with severe AKI were randomly assigned to accelerated-strategy group and standard-strategy group. The primary and secondary outcomes were comparable between the two groups, while more adverse events occurred in the accelerated-strategy group [
14].
Based on a sufficient number of high-quality RCTs, we conducted this systematic review of RCTs with meta-analysis and trial sequential analysis (TSA) to compare the effects of early RRT initiation versus delayed RRT initiation.
Methods
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA statement) guidelines to perform this meta-analysis [
15] (see Additional file
1). No prospectively registered protocol was existed; however, search terms, data extraction, inclusion and exclusion criteria, and data synthesis were according to a plan made by our team.
Eligibility criteria
The inclusion criteria were as follows: (1) population: critically ill patients with AKI aged 18 years or older; (2) intervention: the treatment group received early RRT; (3) Comparison intervention: the control group received delayed RRT; (4) outcome: 28-day all-cause mortality, 90-day mortality, or hospital all-cause mortality were available; and (5) study design: RCT. The exclusion criteria were as follows: (1) study type was not RCT; (2) patients included children; (3) study not focused on critical illness; (4) without a clearly definition of “early” and “delayed” strategies; and (5) the reason for initiating RRT was not AKI, but others. There were no restrictions on publication language.
Search strategy and selection process
We searched PubMed, EMBASE, and the Cochrane Central Register of Controlled Trials Library database from inception through to July 20, 2020. We used key-words and free-text words which were related to AKI, RRT, critical illness and timing of initiating RRT. The detail of search strategy for PubMed is shown in Additional file
2. The reference lists of the included studies and recent review articles were hand-searched to find additional citations. Two authors (X.L and C.L) independently screened all potentially relevant citations to find studies for the final analysis. Any disagreements between two authors were resolved through discussion.
Data extraction and risk of bias assessment
Two authors (X.L and C.L) extracted the following information in a standard form independently: the first author, study center (single-center or multicenter trial), publication year, patient characteristics (mean age of the patient, sample size, male percentage and patient population), details of RRT (criteria for RRT initiation and RRT modality), all clinical outcomes. Two authors (X.L and C.L) independently evaluated the risk of bias for each of these studies by the Cochrane risk of bias assessment tool [
16]. Any disagreements were resolved by discussion, if no agreement could be reached, it would be decided by a third author (F.Z). Only when all the items were assessed as low risk bias, the study was classified as low risk of bias, otherwise the study would be considered as high risk of bias.
Outcomes
The primary outcome was 28-day all-cause mortality. The secondary outcomes included 90-day all-cause mortality, hospital all-cause mortality, ICU all-cause mortality, number of patients who received RRT, RRT dependence at 28-day among survivors, RRT dependence at 90-day among survivors, length of hospital stay, length of ICU stay, mechanical ventilation-free days up to day 28, RRT-free days up to day 28, and vasopressor-free days up to day 28. The incidence of adverse events potentially associated with RRT was also evaluated, including hypotension, any arrhythmia, bleeding events, and infection during the treatment.
Statistical analysis
For binary outcomes, we calculated the risk ratios (RRs) and 95% confidence intervals (CIs) by the Mantel–Haenszel method. For continuous outcomes, we used the inverse variance method to pool the mean differences (MDs) and 95% CIs. Heterogeneity among the included studies was assessed using the
I2 statistic, which the
I2 values of 25%, 50%, and 75% represented low, moderate, and high heterogeneity, respectively [
17]. When the
I2 values < 25%, we used the fixed-effect mode. Otherwise, the random effects model was used as appropriate. If a two-sided P value was less than 0.05, the results were considered statistically significant. We used funnel plots to assess the publication bias [
18]. Subgroup analyses for the primary outcome were performed based on mean age of patients in each study (> 65 years or ≤ 65 years), the SOFA scores at administration (> 12 or ≤ 12), and the criteria for early RRT initiation (Approximately equal to stage 2 of the KDIGO classification, approximately equal to stage 3 of the KDIGO classification, or other classification criteria subgroup) [
19]. We did sensitivity analyses for the primary outcome according to publish language (excluding the study published in Chinese), risk of bias (only including studies classified as low risk of bias), and publish year (removing studies published before 2010). All statistical analyses were performed by Review Manager (version 5.3).
Trial sequential analysis
We conducted TSA to control the risk of random errors and assess whether the results in our meta-analysis were conclusive. We used a random effects model to construct the cumulative
Z curve. TSA was performed to maintain an overall 5% risk of a type I error. Based on previous high-quality RCTs on this topic [
14,
20], we used an anticipated relative risk reduction (RRR) of 15.0% with a power of 90% to calculate the required information size to detect or reject an intervention effect. And the control event rate was adjusted according to the relevant rate of standard therapy (delayed-strategy) group in our meta-analysis. When the cumulative
Z curve crossed the trial sequential monitoring boundary or entered the futility area, a sufficient level of evidence for accepting or rejecting the anticipated intervention effect may have been reached, and no further studies were needed. If the
Z curve did not cross any of the boundaries, and the required information size had not been reached, evidence to reach a conclusion was insufficient, and more studies would be required [
21].
Discussion
This systematic review and meta-analysis included 11 studies comparing delayed versus early initiation of RRT for AKI in critically ill patients. The pooled results showed that early initiation of RRT was not associated with survival benefit in critically ill patients with AKI. The TSA results indicated a RRR of 15% or greater could be rejected with respect to 28-day, 90-day, or hospital mortality. In addition, early initiation of RRT could lead to unnecessary RRT exposure in some patients, resulting in a higher incidence of RRT-associated adverse events, including hypotension and infection.
Over the past few decades, RRT has become more sophisticated, with more modalities available, each with its own merits in particular situations [
8]. RRT can be life-saving by correcting metabolic disorders in patients with severe acidosis and hyperkalemia, controlling disturbances of fluid metabolism in patients with severe pulmonary edema, and removing toxins and circulating inflammatory cytokines in patients with severe sepsis. We can learn from the inclusion criteria for each of the included studies that two studies included patients with sepsis, one study included patients with shock after cardiac surgery, and the other eight studies included mixed populations (see Additional file
7). Causes and pathophysiological mechanisms of AKI were highly variable in different studies, such as renal hypoperfusion, nephrotoxin exposure, ischemic reperfusion injury, and an increase in the level of circulating inflammatory cytokines. To our knowledge, the prognosis of RRT for AKI induced by different causes may be different. Moreover, the criteria for the initiation of RRT, the definition of AKI and RRT modalities existed great variations among the included studies. Therefore, we should be cautious with the results of this study.
So far, this is the only meta-analysis including the STARRT‑AKI trial [
14]. Our results were consistent with the results of most previous meta-analyses except three, which reported early initiation of RRT may have significant benefit on survival [
9,
31,
32]. However, a considerable proportion of the included studies in these three reviews were non-RCTs, meaning that the data were prone to confounding factors. Another two meta-analyses with TSA of RCTs conducted by Moreira et al. and Feng et al. failed to establish sufficient and conclusive evidences, because the cumulative
Z curve did not cross the conventional boundary, the trial sequential monitoring boundary and the futility boundary, and the required information size was not reached [
33,
34]. However, in our meta-analysis, the cumulative
Z curve crossed the futility boundaries, suggesting the results that early initiation of RRT was not associated with a lower mortality were reliable.
Hemodynamic instability is a common complication during RRT, which can increase hospital mortality and limit kidney recovery [
35,
36]. Many factors contribute to hemodynamic instability, including excessive ultrafiltration, rapid osmotic/oncotic shifts, decreased cardiac output, and decreased peripheral resistance [
37]. The incidence of hypotension was 15.7% and 11.0% in the early-strategy group and in the delayed-strategy group, respectively. We can learn from Table
1 that studies which reported hypotension events all involved intermittent hemodialysis (IHD), which was more likely to result in hemodynamic instability than CRRT. A significant difference detected between the two groups may be due to more patients in the early-strategy group exposure to RRT (2468 of 2539 patients) compared with the delayed-strategy group (1591 of 2547 patients). However, none of the included studies reported the mode of RRT when the hypotension occurred. Therefore, we failed to find the association between RRT mode and hypotension in this meta-analysis. There were no statistical significances in hospital mortality and kidney recovery between the two groups, but hospital mortality and 90-day RRT dependence rates were higher in the early RRT group than the delayed RRT group. A remarkable higher incidence of RRT-associated infection events was also found in the early RRT group. Patients treated with RRT are more susceptible to infection, as they are exposure to catheters and invasive treatments [
38,
39]. Moreover, RRT may enhance the elimination of antibiotics, leading to suboptimal antibiotic concentrations [
40].
In this meta-analysis, only 62.5% patients in the delayed-strategy group received RRT. Although fewer patients received RRT in the delayed-strategy group compared to the early-strategy group, the clinical outcomes were comparable between the two groups. In addition, our results showed that delayed RRT initiation could reduce the incidence of RRT-associated adverse events. Undoubtedly, unnecessary RRT will increase the workload of medical staff, augment treatment costs, and waste health resources. Therefore, it is reasonable to assume that delayed initiation of RRT is a preferable approach for critically ill patients with AKI.
As shown in Table
1, the criteria for initiating RRT and definition of AKI were associated with great variations among the included studies. Timing of RRT initiation was determined by AKI stage, biochemical marker level or urine output. AKI was defined by the RIFLE (risk, injury, failure, loss, and end-stage) criteria, the AKIN (AKI Network) criteria, or the KDIGO (Kidney Disease: Improving Global Outcomes) criteria. One problem was that patients who were classified into the early-strategy group in one study might be classified into the delayed-strategy group in another. Despite the definition of early RRT had differences among the included studies, the criteria for early RRT initiation were similar in some studies. And the subgroup analyses based on the criteria for early RRT initiation also showed that early RRT could not decrease 28-day all-cause mortality compared with delayed RRT. Although there were differences in the definition of the delayed RRT, most of studies initiated RRT when patients were with severe complications such as severe pulmonary edema, severe acidosis, and severe hyperkalemia. It is reasonable for us to assume that the optimal timing of initiating RRT is when patients are with severe complications. It is also worth noting that although there are a variety of criteria for initiating RRT, it is mainly based on renal function indicators at present. RRT cannot only influence renal function, but also have an effect on other organs, such as liver function, cardiac function, and so on [
41,
42]. Perhaps establishing a scoring system based on systemic multi-organ functions to find the best cutoff time initiating RRT is the way forward, just like Sequential Organ Failure Assessment (SOFA) score and Acute Physiology and Chronic Health Evaluation II (APACHE II) score.
The strengths of our study are as follows: First, we only included RCTs and most of the included studies were assessed as low risk of bias. Second, we comprehensively evaluated the effect of RRT initiation timing on clinical outcomes, including mortality, renal function recovery, various adverse events, and so on. Third, we performed TSA to determine whether the evidences in our research were reliable. Notwithstanding the aforementioned, there are several limitations in our study. The main limitation is the criteria for the initiation of RRT had great variations among the included studies. Second, we did not perform subgroup analyses according to RRT modalities, delivered dialysis dose. The choice of the RRT modality in most included studies were prescribed and monitored according to national guidelines. Some patients received CRRT at the outset, but may switch to other RRT modalities depending to their conditions. We tried to find whether the choice of RRT modality may influence the results. However, since this was a secondary analysis study, the individual patient data was not available. We cannot further analysis the effect of RRT modality on outcomes.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.